Improving It Service Desk Through the Use of Ai-powered Itsm Chatbot: a Case Study of Moh Kenya Service Desk

Abstract

Health information systems ensure the timely delivery of care to patients. The Ministry of Health of Kenya through its medium-term plan for Vision 2030, identified the need to strengthen national health information systems with timely and understandable information on health. The USAID HealthIT Project is one such initiative that’s working towards attainment of this goal. It aims to bolster evidence-based decisioning among all stakeholders in the health sector. It does so by reinforcing integrated Health Information Systems (HIS) to enhance their effectiveness. In this regard it has assisted MOH in the creation and maintenance of user service desk, with the aim of provision of efficient, effective, and timely first line of support to help resolve ICT and HIS system issues within the health sector space. Advances in Artificial Intelligence (AI) can be leveraged to automate the service desk, which can lead to timely delivery of the much-needed patient care. The primary aim of this research is to enhance the service desk by leveraging Natural Language Processing (NLP) and deep learning techniques. The methodology employed in the research is action research, where a proof-of-concept bot is developed using Rasa framework with select support ticket classes as the bot capabilities. The resulting model was then integrated with WhatsApp, and through the help of subject matter experts the usefulness and effectiveness of the bot are evaluated. The developed handled four ticket classes i.e., password reset, system account creation, network issue troubleshooting, and display problem troubleshooting. The resulting model was evaluated using f1-score, precision, and recall, with each having a score of 1.0 for each of the classes. The model over-fitted due to the amount of training data used for the training classes. The evaluation of the bot done via a group discussion with subject matter experts, indicates that if the bot is deployed to production, it could reduce the number of support cases by close to 30 percent. This is based on their experience and not on concrete turnaround numbers which can only be measured if the system is deployed and evaluated after some usage time.